Issues in Implementing Problem–based Learning Online

David Jonassen

Issues in Problem-based Learning

Problem-based learning (PBL) is an instructional methodology. The PBL model calls for the construction of problem sets of authentic problems and the engagement of learning groups in negotiating learning issues in order to solve those problems. The problem-solving activity leads to learning. In face-to-face classes, PBL normally involves the following process:

1. Students in groups of 5-8 encounter and reason through the problem. They attempt to define and bound the problem and set learning goals by identifying what they know already, what hypotheses or conjectures they can think of, what they need to learn in order to better understand the dimensions of the problem, and what learning activities are required and who will perform them.
2. During self-directed study, students individually complete their learning assignments. They collect and study resources and prepare reports to present to their own group.
3. Students share their learning with the group and revisit the problem, generating additional hypotheses and rejecting unlikely ones based on their learning.
4. At the end of the learning period (usually one week), students summarize and integrate their learning.

Although PBL has been shown to be successful in supporting deep-level and lifelong learning in biomedicine, law, and, to lesser degrees, business and engineering (Hung et al., in press), a number of issues must be resolved if we are to successfully implement PBL in online environments. In this chapter, I will briefly address three (of many) issues:

• What kinds of problem solving can be effectively supported with PBL?
• How can PBL methods be scaled to work effectively online?
• How can we effectively tutor PBL online?

PBL Meets Problem Solving

The PBL methodology assumes that all problems are solved in the same way and, therefore, learning to solve any problem requires the same processes. My research on problem solving questions that assumption. The nature of problems is determined by external and internal factors. External factors include the structuredness (reliability) of the problem (Jonassen, 1997). The most commonly encountered problems, especially in schools and universities, are well-structured in nature. Well-structured problems typically present all the elements of the problem, engage a limited number of rules and principles that are organized in a predictive and prescriptive arrangement, possess definite and convergent answers, and have a preferred and prescribed solution process. Ill-structured problems, on the other hand, are the kind that is encountered in everyday practice. They have many alternative solutions, vaguely defined or unclear goals and constraints, multiple solution paths, and multiple criteria for evaluating solutions. Hence, they are more difficult to solve.

External factors of problems also include the complexity of the problem, the dynamicity of the problem, and the role of problem context. Complexity is a function of the difficulty that problem solvers experience as they interact with the problem, the importance of the problem (the degree to which the problem is significant and meaningful to the problem solver), and the urgency to solve it (how soon the problem should be solved). Complexity is described by the number of factors involved and their interactions. It is also a function of the dynamicity of the problem, that is, the degree to which problem factors or elements change over time. The context in which problems occur can also make them more complex. In everyday problems, which tend to be more ill-structured, context plays a much more significant role in the cognitive activities engaged by the problem (Lave, 1988; Rogoff & Lave, 1984). The context in which the problem is embedded becomes a significant part of the problem and, necessarily, part of its solution (Wood, 1983).

Internal factors relate to the characteristics of the students who are learning to solve problems. Intellectual dispositions, such as intelligence, cognitive skills, epistemological beliefs, and domain knowledge will have significant impact on the complexity and difficulty of solving the problem. More important, however, is the level of problem-solving experience that the learners possess. Because of the diversity of individual differences, it is practically impossible to adapt the nature of instructional support to meet the needs of individual learners. Therefore, PBL environments must focus on supporting student interaction through structuring the external problem factors.

Based on external factors, Jonassen (2000) describes a typology of problems along a continuum from well-structured to ill-structured, including algorithms, story problems, rule-using problems, decision making, troubleshooting, diagnosis-solution problems, strategic performance, systems analysis, design problems, and dilemmas. This typology assumes that there are similarities in the cognitive processing engaged within these classes of problems. Within each category of problems that is described, problems can vary with regard to structuredness, complexity, dynamicity, and context. Because of space limitations, I will describe only some of these problems.

Story problems
Story problems are the most commonly encountered type in formal education. From elementary mathematics through graduate-level dynamics, textbook chapters present story problems. Story problems include elementary combine, cause/change, or compare problems in mathematics (Joe has three marbles, Jane gave him three more, how many does he have?); calculating resistance given voltage and amperage; calculating the amount of reagents needed to form a specific precipitate in a chemical reaction; or calculating the interest accrued on a savings account. They usually take the form of a brief narrative or scenario in which the values needed to solve an algorithm are embedded. In order to meaningfully solve story problems, learners must construct and use a problem schema and apply that schema to the current problem. If they use the correct schema, the solution procedure is usually embedded within that schema. However, numerous difficulties can occur when students extract the values from the narrative, insert them into the correct formula, and solve for the unknown quantity, because they focus too closely on the surface features of the problems (Woods et al., 1997). Why do students use this ineffectual strategy? Because too often that is how they have been taught to solve problems. Teachers and textbooks demonstrate the procedures for identifying variables, inserting the variables into formulae, and solving for the unknown.

In the case of online courses, a similar tutorial strategy is typically used to teach the solution of story problems. Simple feedback is provided, but students receive no conceptual instruction. Solving story problems requires understanding of the structure of the problems. The instruction must emphasize the structural components of the problem and its context and distinguish the structure of any problem from all others in the domain (Jonassen, 2003).

Rule-using problems
Many problems have definite solutions but multiple solution methods and multiple rules governing the solution process. They tend to have a clear purpose or goal that is constrained but not restricted to a specific procedure or method. Rule-using problems can be as simple as expanding a recipe to accommodate more guests and as complex as completing tax return schedules or evaluating a loan application. Using an online search system to locate relevant information on the World Wide Web is a good example of rule-using problems. The purpose is clear: find the most relevant information in the least amount of time. That requires the selection of search terms, construction of effective search arguments, implementation of the search strategy, and evaluation of the utility and credibility of the information found. Although research on online searching abounds, there is none that conceives of it as a kind of problem solving. Learning how to apply multiple rules to problems in a conceptual way, both face to face and online, needs to be studied.

Diagnosis problems
Diagnosis (also known as troubleshooting) is one of the most commonly accepted forms of everyday problem solving. Maintaining automobiles, aircraft, or any complex systems requires diagnostic skills. Debugging computer programs requires diagnosis. Diseases have to be diagnosed before they can be treated. In situations requiring diagnosis, some part or parts of a system are in an abnormal state, resulting in a set of symptoms that have to be matched with the user’s knowledge of various fault states. Diagnosticians use symptoms to generate and test hypotheses about different fault states. Unfortunately, diagnosis is too often taught as a procedure. In addition to procedural knowledge of the diagnostic process, troubleshooting also requires conceptual knowledge of the system, strategic knowledge of when to use tests and methods, and causal reasoning for hypothesis generation and testing. Later in this chapter, I will briefly demonstrate an architecture for representing and supporting meaningful troubleshooting online.

Strategic performance
Strategic performance entails complex activity structures in a real-time environment where the performers apply a number of tactical activities to meet a more complex and ill-structured strategy while maintaining situational awareness. In order to achieve the strategic objective of, say, teaching in a classroom or quarterbacking a professional football offense, the performer applies a set of complex tactical activities that are designed to meet that objective. Pursuing strategies through tactical activities requires the application of a finite number of such activities that have been designed to accomplish the strategy. However, an expert tactical performer is able to improvise or construct new tactics on the spot to meet the strategic objective, such as devising a new strategy in a courtroom when unexpected evidence emerges. Those adjustments are contextually constrained. In traditional education, students are usually taught the prerequisite skills but not how to engage and regulate them in real time. That skill requires real-time simulations, which are difficult to provide in both face-to-face and online instruction.

Design problems
Among the most ill-structured but meaningful problems are design problems. Whether designing instruction, an electronic circuit, a race car, a marketing campaign for a new Internet company, or any other product or system, designing requires the application of a great deal of domain knowledge with considerable strategic knowledge to produce an original design. What makes design problems so ill-structured is that there are seldom clear criteria for evaluating success. The client either likes or dislikes the result but seldom can articulate why. Therefore, skills in argumentation and justification help designers to rationalize their designs. Although designers always hope for the best solution, the best solution is seldom ever known. Also, most design problems are complex, requiring the designer to balance many needs and constraints in the design and of the client. Despite the difficulties, design problems are among the most common type encountered in professional practice. Virtually every engineer, for example, is paid to design products, systems, or processes.

PBL Support for Problem Types

An important research question concerns the nature of problems that are amenable to PBL. PBL emerged in medical schools with students learning to solve diagnosis-solution problems, which are moderately ill-structured. The goal of medical diagnosis is to find the source of any physiological anomaly. However, there are numerous paths that can lead to a diagnosis. And in the treatment or management part of the solution process, the problem often becomes more ill-structured because of multiple treatment options, patients’ beliefs and desires, insurance coverage, and so on. PBL has since migrated to law schools, where students learn to construct arguments based on evidentiary reasoning, a complex form of rule-using problem. PBL is becoming increasingly popular in graduate business programs, where students primarily solve case analysis problems that are fairly ill-structured. As PBL migrates to other academic programs, such as engineering, the question about the kinds of problems that are amenable to PBL environments will become more important.

Along the continuum from well-structured to ill-structured problems, which kinds of problems can be effectively supported using PBL? How well-structured or ill-structured can the problems be that would allow self-learning through PBL? How complex can the problems be? For example, can PBL be adapted to story problems in physics, despite the inauthentic nature of those problems? The most common kind of problem in engineering is the design problem, which typically tends to be the most complex and ill-structured kind of problem that can be solved. Given an initial statement of need, there are an infinite number of potential solutions. Can learners self-direct their ability to solve this kind of problem, or is some form of studio course required to accommodate its complexity? What is the range of complexity and structuredness that can be effectively learned using PBL? This will require comparison of the successes and failures of problem solving across domains.

I believe without empirical support that PBL is most applicable to moderately ill-structured problems that require constrained forms of reasoning. PBL will be less effective for story problems in the sciences because the problems are not authentic and complex enough, and the reasoning processes are too limited. Most of us have memories of solving uninspiring mathematics story problems, so it would be difficult to convince students to self-direct their learning using these problems. At the other end of the continuum, very ill-structured problems, like designs and dilemmas, are too complex with too many (or no) acceptable solutions to be amenable to PBL. Design problems require a studio approach to learning. This leaves diagnosis, strategic performance, and systems analysis problems as the most amenable to PBL approaches. Clearly, research is needed to confirm my hypothesis.

Scaling PBL in Online Environments

With the emergence of online learning initiatives, researchers are working to implement PBL in online environments. This trend raises numerous implementation issues. How faithfully can PBL methodologies be applied online? How can learning groups collaborate effectively online in order to negotiate meaning? How can tutors effectively nurture and guide learning online? How can we support self-directed learning online? What compromises, if any, are required to engage learners in PBL online?

To customize PBL environments to support different kinds of problems in every different domain and context will be prohibitively expensive. The only way, I believe, that PBL can be scaled to function effectively in online environments is to design and develop problem-specific architectures or authoring environments. I next describe such an architecture for diagnosis problems. As noted earlier, effective diagnosis requires learners to construct system knowledge, procedural knowledge, and strategic knowledge. Those kinds of knowledge are anchored in and organized by the diagnoser’s experiences. The biggest difference between expert and novice diagnosticians is their level of experience. That is, the ability to diagnose relies on experiential knowledge, which is exactly what novices lack. Experienced physicians construct “illness scripts,” which they use to diagnose the illnesses of new patients (Schmidt et al., 1990). They have a rich store of past patients’ histories and diagnoses which they have used to build these scripts. They then compare new patients’ conditions to specific scripts in order to make a diagnosis, using pathophysiological reasoning only when they encounter a patient with a new set of symptoms that they have not encountered before. Thus, novices learning to diagnose need to diagnose as many problems as possible in order to gain the experiential knowledge that integrates conceptual, procedural, and strategic knowledge.

Figure 11.1 illustrates a design model for building diagnosis learning environments (DLEs). Learning to diagnose problems requires that learners practice solving novel problems based on the symptoms given. The major components of the DLE are a multilayered conceptual model of the system being diagnosed, a simulator that enables the learner to practice diagnosis, and a case library of previously solved problems. The conceptual model supports the construction of system knowledge, the simulator supports the construction of procedural and strategic knowledge, and the case library supports the construction of the experiential knowledge that integrates all of the other kinds of knowledge.

Conceptual Model

The DLE is oriented by the conceptual model of the system being diagnosed. The conceptual model illustrates the interconnectedness of system components. It provides multiple layers of system representation, including the following:

• The pictorial layer contains pictures of the system as it exists.
• The topographic layer illustrates the components of the system, their locations, and their interconnections. Topographic representations of the system are important because experts rely on them to search for faulty components (Johnson, 1988; Rasmussen, 1984).
• The state layer provides several overlays to the topographic layer. Each overlay conveys the normal states or values of each component.
• The symptom layer conveys the symptoms associated with each component malfunction. Matching existing symptoms and probabilities with a set of symptoms and probable fault states stored in the system database represents a common approach to fault finding (Patrick, 1993).
• The functional layer illustrates and describes the information, energy, or product flows through the system and how the components affect each other. Understanding system functions is more effective than understanding strategic advice in fault finding (Patrick & Haines, 1988), but the combination should be even more effective.
• The probability layer conveys the probabilities of the fault states.
• The strategic layer consists of rule-based representations of alternative decisions regarding the states described on the state layer. It contains diagnostic heuristics that support fault finding (Patrick & Haines, 1988).
• The action layer includes descriptions of procedures for conducting various tests or operations (a job aid or just-in-time instruction for students performing various tests or other actions).

An integrated conceptual understanding of the system is essential for diagnosis. Layers of the model provided can be accessed from anywhere in the DLE on demand.

Simulator

The heart of the DLE is the diagnoser or troubleshooter. This is where the student learns diagnostic reasoning. In a medical DLE, for example, after reading a patient’s medical history and viewing a videotape of the physical examination, the learner first selects an action from a pull-down menu, such as ordering a test or taking blood pressure. The novice may be coached by an animated pedagogical agent (described later) about what action to take or what issue to think about. Each action taken by the learner shows up in the system model. For each action the learner takes, the simulator next requires the learner to state or select from a pull-down menu a hypothesis that he or she is testing. This is an implicit form of argumentation (detailed later) requiring the learner to justify the action taken. If the hypothesis is inconsistent with the action, then feedback can be immediately provided on the rationale for taking such an action. Next, the learner must also identify the subsystem in which the fault occurs. If the subsystem identified is inconsistent with the action, the learner is immediately sent to the conceptual model to better understand the workings of the correct subsystem that leads to the action or hypothesis. The learner then receives the result of the action (e.g., test results) and must interpret those results with the aid of a pull-down menu. If the interpretation is inconsistent with the action, hypothesis, or subsystem, then an error message is triggered. The error checking uses a very simple evaluation system.

Case Library

If the troubleshooter is the heart of the DLE, the case library is the head (or memory). When diagnosing problems in everyday contexts, the primary medium of negotiation is stories. That is, when a problem is experienced, the troubleshooter first tries to recall a similar problem that he or she has solved previously and its solution. If none comes to mind, the troubleshooter most often will describe the problem to someone else, who may then recall a similar problem and relate it to the troubleshooter. These stories provide contextual information, work as a format for diagnosis, and also forge an identity among participants in a practice community. Stories about how experienced troubleshooters solved similar diagnosis problems are contained in, indexed by, and made available to learners in a case library.

The case library should contain stories of as many trouble-shooting experiences as possible. Each case represents a story of a domain-specific diagnostic instance. Case libraries, based on principles of case-based reasoning, represent the most powerful form of instructional support for ill-structured problems such as troubleshooting (Jonassen & Hernandez-Serrano, 2002). The case library indexes each case or story according to the following features:

• Symptoms observed
• Actions or procedures required to isolate the faults
• Hypothesis tested
• Results of various tests
• Topographic component
• Specific disease state
• Solution strategies
• Other relevant aspects of the case

The case library represents the experiential knowledge of potentially hundreds of experienced physicians. So, when learners encounter any difficulty or are uncertain about how to proceed, they may access the case library to learn about similar cases, what was done, and what the results were. The DLE can also be programmed to automatically access a relevant story when a learner commits an error, orders an inappropriate test, or takes some other action that indicates a lack of understanding. Stories can be easily collected from experienced physicians by presenting them with a problem and asking them if they are reminded of a similar problem that they have solved. Invariably they are. Hernandez-Serrano and Jonassen (2003) have shown that accessing a case library when learning how to solve problems improves complex problem-solving performance in an examination.

Learners are introduced to the simulator, case library, and conceptual model through worked examples that not only illustrate how to use the DLE but also model different troubleshooting strategies (e.g., space splitting) for isolating the faulty components. We have been exploring the use of animated pedagogical agents to tutor learners by providing metacognitive prompts when needed and giving feedback related to learner performance. The agent demonstrates through at least two examples of each problem type. It reads the problem representation in the DLE and models strategies such as looking for clues, rejecting the least likely hypothesis, and other strategies before turning to the troubleshooter. The agent also models how to gain the most benefit from the conceptual model and the case library.

I have briefly described an architecture for designing DLEs. What is needed for adapting and scaling PBL to online environments are authoring environments based on this architecture that support the collection of stories, the analysis of diagnostic reasoning, and the representation of the conceptual model. An advantage of this approach is that the conceptual model and the case library can be developed just once and used to support as many problems as you want to embed in the simulator. Besides this diagnostic architecture, we have also developed architectures for story problems and systems analysis problems, and we are working on one for design problems. The efficacy of this approach must be established through large-scale testing.

Tutoring PBL Online: Scaffolding Negotiation

The power of PBL lies in the social negotiation of meaning that occurs in study groups. As students collaboratively articulate what they know about the problem being solved, generate hypotheses and diagnoses, and identify the gap between what they know and what they need to learn, they are engaged in a highly specific form of self-directed learning. These face-to-face negotiations are coached by a tutor, who supports and models diagnostic processes without interjecting content, facilitates group processes, and monitors individual learning. The discursive interactions between students are purposeful and intensive as students share their understanding, make conjectures about the problem, and self-evaluate their understanding related to the case.

Among the vexing problems related to conducting PBL online is how to engage and support the complex negotiations of study groups. A primary issue is whether to conduct negotiations synchronously or asynchronously. Negotiations are carried out synchronously in face-to-face classes, but scheduling online chat may be difficult if students are geographically dispersed. Chat would also have to be supported with shared whiteboards and other negotiation tools. I briefly describe some solutions that I have worked on.

Scaffolding Conversation

When collaborating to solve problems, especially ill-structured ones, justification and argumentation are more important than they are for solving well-structured problems (Shinn et al., 2003). As authentic problems are ill-structured and complex, implementing a solution for assessing performance would require too much time and effort. And as ill-structured problems possess multiple solutions and no convergent solution path, a primary form of assessment of problem-solving skills is the ability to justify or argue for the solution that one generates. Argumentation is a fundamental process of social negotiation through informal reasoning. Negotiating online does not readily expose the informal reasoning processes, which are important to problem solving. In order to support social negotiation, it is essential to make this informal reasoning explicit (Senge, 1990).

In my research, I have examined the use of constraint-based discussion boards for scaffolding different kinds of reasoning (Jonassen & Remidez, 2005; Oh & Jonassen, in press). Constraint-based discussion boards use prestructured forms of conversation systems that impose different rhetorical structures onto the discussion. These structures constrain the conversation that students conduct. Preclassifying conversational attributes to fit these sets of canonical relations constrains the nature of verbal interactions among conversers. For example, using C-Board, developed at the University of Missouri, instructors can construct a set of rhetorical constraints for structuring conversation. For a discussion board in which students are collaboratively diagnosing a patient’s condition, students could, for example, start a new discussion thread only with a hypothesis. They would type in their diagnosis (e.g., bladder infection). Any students wishing to respond to that hypothesis may provide warrants for supporting that diagnosis or rebutting it. Any students wishing to elaborate on those warrants would be constrained to provide different forms of case evidence that support either of those warrants (e.g., physical examination data, family history, laboratory test results). The conversation constraints model the kind of reasoning that students must learn to perform, preventing them from making irrelevant responses. Students’ responses provide valuable information for assessing their understanding as well as for evaluating the processes that they are using. Cho and Jonassen (2002) showed that such constraints were more effective when learning to solve ill-structured problems than with well-structured problems.

A wide variety of conversational constraints may be used for scaffolding different kinds of reasoning. A discussion board focused on evaluating alternative solutions might be scaffolded by multiple structures that would examine each solution in terms of cost, feasibility, time, and effort, each of which could be further supported with case evidence, such as the different perspectives of the attending physician, the patient, the surgeon, the physiologist, and others. Such discussion constraints would support various forms of role-playing by students in the class.

Scaffolding Reasoning

Different forms of reasoning may also be constrained by the task structure in the environment. In other words, as students work through an online process, the environment engages and supports specific kinds of reasoning. For example, in designing a medical DLE, we developed a causal model of every etiology and differential diagnosis related to platelet diseases (Jonassen et al., 1996). Causal modeling provides a structure for elaborating clinical events that leads learners to the causes and effects (e.g., what signs and symptoms point to a particular etiology), thereby providing a scenario for learners. Based on that model and specific case findings, learners are required to support each etiology and diagnosis (initial and differential) along with every treatment prescription with specific case evidence (see Figure 11.2). Students must identify how important different case findings are for supporting an initial diagnosis. Discrepancy scores between students’ choices and the expert’s are reported to the instructor for grading. These scores also serve as feedback to help students determine if their diagnosis should be altered. This task structure supports the kind of premise-evidence reasoning that is required to make medical diagnosis.

In a similar PBL environment where students are learning about the broad range of nuclear sciences by solving problems related to elemental and content analysis, material modification, radiation gauging, radiation imaging, and nuclear power, students must determine the radiation source and detection methods required to solve the problem and then justify their answers using information provided on the course web site (Jonassen et al., 2005). Figure 11.3 illustrates a radiation-gauging problem where students are required to select the best radioisotope for projecting through the sheet of aluminum to gauge its thickness and then justify their response. In subsequent screens, learners argue for the detector they decide to use. Any form of reasoning or argumentation may be embedded in the task structure of the online learning environment. Without a readily available tutor, the PBL environment must assume more responsibility for supporting student thinking.

Summary

The migration of all forms of learning to online environments seems inevitable. That migration will definitely require adaptation of various pedagogies in order to support the different needs of online learners. In this chapter, I have raised some important questions about how to adapt PBL for online implementation, including the kinds of problem solving that can be effectively supported with PBL, how to adapt and

From the following list of isotopes, select the isotope that will provide adequate energy for gauging thickness for the longest period of time.

• Americium (Am-241)
• Cesium (Cs-137)
• Cobalt (Co-60)
• Strontium (Sr-90)
• Krypton (Kr-85)

4.1 What isotope from the table above would you recommend if half-life was not a factor in your choice? Discuss your choice in terms of:

• Type of energy emitted
• Attenuation

scale PBL methods, and how to scaffold PBL processes online. I do not presume that my recommendations are the only reasonable ones. Nor do I presume that I have addressed all of the issues. I offer this discussion and my recommendations as a starting point for discussion and experimentation.

References

Cho, K. L., & Jonassen, D. H. (2002). The effects of argumentation scaffolds on argumentation and problem solving. Educational Technology Research and Development, 50 (3), 5-22.

Hernandez-Serrano, J., & Jonassen, D. H. (2003). The effects of case libraries on problem solving. Journal of Computer-Assisted Learning, 19, 103-14.

Hung, W., Jonassen, D. H., & Liu, R. (in press). Problem-based learning. In J. M. Spector, J. G. van Merrienboer, M. D., Merrill, & M. Driscoll (eds), Handbook of research on educational communications and technology, 3rd Ed. Mahwah, NJ: Lawrence Erlbaum Associates.

Johnson, W. B. (1988). Developing expert system knowledge bases in technical training environments. In J. Psotka, L. D. Massey & S. A. Mutter (Eds.), Intelligent tutoring systems: Lessons learned (pp. 21-33). Hillsdale, NJ: Erlbaum.

Jonassen, D. H. (1997) Instructional design model for well-structured and ill-structured problem-solving learning outcomes. Educational Technology: Research and Development 45 (1), 65-95.

Jonassen, D. H. (2000). Toward a design theory of problem solving. Educational Technology Research and Development, 48 (4), 63-85.

Jonassen, D. H. (2003). Designing research-based instruction for story problems. Educational Psychology Review, 15 (3), 267-296.

Jonassen, D. H., & Hernandez-Serrano, J. (2002). Case-based reasoning and instructional design: Using stories to support problem solving. Educational Technology Research and Development, 50 (2), 65-77.

Jonassen, D. H., & Remidez, H. (2005). Mapping alternative discourse structures onto computer conferences. International Journal of Knowledge and Learning, 1 (1), 113-29.

Jonassen, D. H., Mann, E., & Ambruso, D. J. (1996). Causal modeling for structuring case-based learning environments. Intelligent Tutoring Media, 6 (3/4), 103-12.

Jonassen, D. H., Schmidt, M. A., Miller, W., & Neumeyer, G. (2005). A problem-based introduction to nuclear sciences. In Proceedings of the Annual Conference of the American Society of Engineering Education. Washington, DC: ASEE.

Lave, J. (1988). Cognition in practice: Mind, mathematics and culture in everyday life. Cambridge: Cambridge University Press.

Oh, S. C., & Jonassen, D. H. (in press). Scaffolding argumentation during problem solving. Journal of Computer-Assisted Learning.

Patrick, J. (1993). Cognitive aspects of fault-finding training and transfer. Le Travail Humain, 56 (2/3), 187-209.

Patrick, J., & Haines, B. (1988). Training and transfer of fault-finding skill. Ergonomics, 31 (2), 193-210.

Rasmussen, J. (1984). Strategies for state identification and diagnosis in supervisory control tasks, and design of computer-based support systems. In W. B. Rouse (Ed.), Advances in man-machine systems research, Vol. 1 (pp. 139-93). Greenwich, CT: JAI Press.

Rogoff, B., & Lave, J. (1984). Everyday cognition: Its development in social context. Cambridge, MA: Harvard University Press.

Schmidt, H. G., Norman, G. R., Boshuizen, H. P. A. (1990). A cognitive perspective on medical expertise: Theory and implications. Academic Medicine, 65, 611-21.

Senge, P. M. (1990). The fifth discipline: the art and practice of the learning organization. New York: Doubleday.

Shin, N., Jonassen, D. H., & McGee, S. (2003). Predictors of well-structured and ill-structured problem solving in an astronomy simulation. Journal of Research in Science Teaching, 40 (1), 6-33.

Wood, P. K. (1983). Inquiring systems and problem structure: Implications for cognitive development. Human Development, 26, 249-65.

Woods, D. R., Hrymak, A. N., Marshall, R. R., Wood, P. E., Crowe, Hoffman, T. W., Wright, J. D., Taylor, P. A., Woodhouse, K. A., & Bouchard, C. G. K. (1997). Developing problem-solving skills: the McMaster problem solving program. Journal of Engineering Education, 86 (2), 75-92.